Imagine a new receptionist starting on Monday. Nobody tells them the opening hours, whether you cover the next town over, or what to say when someone asks for a discount. By lunchtime they have guessed at three things and apologised for four more.
An AI receptionist starts in exactly that position, except it never asks you a question in the corridor. Whatever you did not teach it, it either admits it does not know or, if the product is badly built, makes something up. So AI receptionist setup is mostly a content job, and it rewards the same preparation you would give a human starter.
Below are 20 things to teach it before day one, grouped the way visitors actually ask them. Tick them off, and where the answer is "we don't have a policy for that", decide one now.
The basics a new starter would ask on their first morning
These are dull, which is why they get skipped. They are also the questions visitors ask most.
- Opening hours, including exceptions. Normal hours, bank holidays, the week you close in August. Say what "open" means if the office and the workshop keep different hours.
- Where you are and where you go. The address, parking, which entrance, and for mobile businesses the towns or postcodes you cover and the ones you do not.
- How to reach a person. Phone number, email, and realistic response times for each. If nobody answers the phone on Fridays, say so.
- Who you serve. Domestic or commercial, adults only, members only, minimum project size. This saves a lot of wasted conversations.
- Your name and what you do, in one plain sentence. Visitors land on deep pages and sometimes are not sure they are on the right site.
- What happens after they get in touch. "We reply within one working day and usually visit within a week" is exactly the sentence nervous customers want.
What you sell and what it costs
- A list of services in the words customers use. "Boiler service" and "annual gas safety check" might be the same job to you but two different searches to a customer.
- What you do not do. A short list of common requests you turn down stops the receptionist from sounding vaguely hopeful about them.
- Prices, or how pricing works. Fixed prices where you have them, ranges where you can, and the factors that change a quote where you cannot give either.
- Lead times. How far ahead you are booked, and whether urgent work jumps the queue at extra cost.
- Payment. Deposits, accepted methods, when payment is due, finance if you offer it.
Policies people ask about before committing
- Cancellations and rescheduling. Notice required, any fee, and how to do it.
- Guarantees and complaints. What you put right, for how long, and who handles a complaint.
- Insurance, accreditations and qualifications. Only the ones you genuinely hold, worded as you would put them on a quote.
- Data and privacy basics. What you do with the details they give you, in a sentence or two, with a pointer to your privacy page.
The awkward cases
This group is where a receptionist earns its keep or embarrasses you.
- Emergencies. What counts as urgent in your trade and exactly what to tell someone out of hours. For anything involving health or safety, the receptionist should give information only and point people to a qualified person or the emergency services.
- Booking requests. What details to collect (name, contact, service, preferred times) and what to say about confirmation. A website receptionist should capture the request and make clear a person will confirm it.
- Discounts and haggling. Whether you ever discount, and the polite line when you do not.
- Existing customers with a problem. Whether to open a support ticket, take details, or give a direct contact.
- What to do when it does not know. The most important item on the list. It should say so honestly, take the visitor's details and pass the question on. More on this in what to do when your agent keeps saying I don't know.
Where each item should live
There are usually several ways to teach an assistant. In SpideyChat that means describing the business in plain language, crawling your website, uploading documents, or adding Q&A pairs. They suit different kinds of fact.
| Kind of information | Best place | Why |
|---|---|---|
| Who you are, who you serve, tone | Plain-language business description | Rarely changes, shapes every answer |
| Service pages, areas, general FAQs | Website crawl | Already written, stays in step with the site |
| Price lists, terms, policy documents | Document upload | Detailed and often not on the site at all |
| Exceptions, haggling lines, holiday closures | Q&A pairs | Short, specific, quick to edit |
If your website is thin on some of this, it is worth fixing the site too, because human visitors need the same facts. The AI FAQ generator can give you a first draft of common questions to fill in, and writing Q&A pairs that sound like you covers how to phrase them.
A worked example of the gaps
Here is an illustration, not a real business. A two-van cleaning company crawls its twelve-page website and asks the receptionist ten questions pulled from last month's inbox.
It answers four well: services, areas, contact details, what end-of-tenancy cleaning includes. Three it answers partly, because the website says "competitive prices" and never gives a figure. Three it cannot answer at all: whether they bring their own products, what happens if a key is not left out, and whether they work Sundays.
Eight Q&A pairs later (three on pricing ranges, one on products, two on access, one on Sundays, one on deposits) it handles all ten. The whole exercise takes about forty minutes, and the owner adds the Sunday answer to the website while she is at it.
Your pre-launch hour
Block an hour and do these in order:
- Pull ten real enquiries from your inbox or voicemail notes, including at least two awkward ones.
- Go through the 20 items above and write the missing answers as short Q&A pairs.
- Ask the receptionist each of the ten enquiries, word for word as the customer wrote them.
- Fix every wrong or vague answer at the source, then ask again.
- Launch on one page first, such as your contact page, and read every conversation for the first week.
After that, the unanswered questions list becomes your checklist for week two, because AI receptionist setup is never really finished. Testing your assistant with real past enquiries goes further on step three, and the first 30 days covers what to watch once it is live.